AI models
TerraMind
Training compute
3.1×10²¹ FLOP
Published
Apr 15, 2025
TerraMind is an AI model developed by IBM, Forschungszentrum Julich, European Space Agency (ESA) and NASA (United States and Germany), first published in April 2025. It works in the earth science and vision domain, on tasks such as image captioning, flood mapping, crop mapping / segmentation and 2 more.
Training it took an estimated 3.1×10²¹ FLOP of compute (estimation method: hardware). It was trained on roughly 500B datapoints. Training ran on 32 NVIDIA A100 for about 288 hours.
Access: Open weights (unrestricted). Its weights are openly available. Epoch AI rates the confidence of this record as confident.
Full record
- Organization
- IBM, Forschungszentrum Julich, European Space Agency (ESA), NASA
- Country of organization
- United States, Germany
- Domain
- Earth science, Vision
- Task
- Image captioning, Flood Mapping, Crop Mapping / Segmentation, Wildfire Mapping, Cloud monitoring / analysis
- Training compute
- 3.1×10²¹ FLOP
- Compute estimation method
- Hardware
- Dataset size
- 500B
- Training hardware
- NVIDIA A100
- Chips used
- 32
- Training time
- 288 h
- Chip-hours
- 9.2K
- Training power draw
- 25.1 kW
- Model accessibility
- Open weights (unrestricted)
- Open weights
- Yes
- Epoch confidence
- Confident
SourceEpoch AI, 'AI Models'. Published online at epoch.ai. Retrieved 2026-07-29 from https://epoch.ai/data/ai-models. Licensed under CC BY 4.0.
← All ai models